fix: strip embedded markup before text conversion

[ump45nose]

What this PR does / why we need it?

markdown_to_plain_text currently applies generic Markdown and HTML cleanup before removing MaxKB's embedded media and form markup. This means:

• the optional closing-tag branch can remove only an <audio> or <video> opening tag and leave fallback content behind;
• Markdown underscore handling rewrites <form_rander> before its JSON payload can be removed.

Those leftovers can reach downstream text-to-speech output.

Summary of your change

• Remove paired audio/video markup, image tags, and form_rander payloads before generic Markdown/HTML conversion.
• Require audio/video closing tags to match their corresponding opening tags.
• Add focused regression coverage for audio fallback text, video fallback text, and form-render JSON.

Validation

• Focused current-source smoke: PASS (audio, video, form-render, and existing Markdown conversion cases)
• ruff check apps/common/utils/common.py apps/common/tests.py
• ruff format --check apps/common/tests.py
• git diff --check

Please indicate you've done the following:

• Made sure tests are passing and test coverage is added if needed.
• Made sure commit message follows the Conventional Commits specification.
• Considered the docs impact; no documentation changes are needed for this bug fix.